For autonomous vehicles to become popular enough, it is necessary to know if they can cope with complex road situations, such as driving on a congested highway. In this sense, researchers from North Carolina State University claim to have developed a technique for unmanned cars to perform calculations faster, which would improve circulation and reduce risk.
“Currently, programs designed to help autonomous vehicles manage lane changes are based on the idea of making problems simple enough for the computer to solve quickly, so that the vehicle can operate in real time. ”, explains Ali Hajbabaie, one of the authors of the study. “However, oversimplifying the problem can actually create new obstacles, as real-life scenarios are rarely straightforward. »
“Our approach allows us to tackle a wide range of ‘real’ world problems. Rather than focus on simplifying obstacles, we designed a cooperative algorithm. This approach essentially breaks a complex problem into several simple sub-problems and sends them to different processors to be processed separately. This process, called parallelization, greatly improves efficiency. »
So far, the researchers have only tested their approach in simulations, where subproblems are shared between different kernels within the same computing system. However, if autonomous vehicles were to use this approach on the road, the cars would connect with each other and share these computing sub-problems, they say.
To assess the feasibility of their solution, the researchers sought to confirm two aspects: first, that their technique actually allows autonomous vehicles to solve lane-change problems in a congested area in real time, and second, to check whether their new “cooperative ” had an impact on traffic and road safety, compared to an existing model that allowed autonomous vehicle navigation.
Regarding computation time, the specialists found that their approach allowed the unmanned vehicles to operate through complex integration scenarios on busy highways, all when the level of congestion was “moderate” or still “high”. However, the effectiveness was reduced when it came to “particularly high” congestion.
However, when it came to improving safety and calming traffic, the new method worked particularly well, the researchers said. Under certain scenarios, especially when congestion was less, both approaches were relatively equally effective. But in most cases, the new approach has proven to be significantly more useful than previous methods. Additionally, the new technique did not cause any time when the vehicles had to come to a complete stop, where they were found to be “in near crash condition.”
The results of the other model include several scenarios where thousands of stoppages and near misses were recorded.
“In terms of theoretical tests, we are very pleased to see how well this technique worked,” says Hajbabaie. “There is still room to improve, but it is a great start. »
“The good news is that we are developing these tools and addressing these issues now, so we will be in a good position to ensure that safe autonomous driving systems are more widely adopted. »